linux-kernel - Re: [PATCH] sched/fair: Do not scan non-movable tasks several times

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <2cf94373-4f25-4a33-a0b4-cab04031bae7@virtuozzo.com>
Date: Mon, 25 Dec 2023 16:17:49 +0100
From: Konstantin Khorenko <khorenko@...tuozzo.com>
To: Vincent Guittot <vincent.guittot@...aro.org>
Cc: Ingo Molnar <mingo@...hat.com>, Peter Zijlstra <peterz@...radead.org>,
 Juri Lelli <juri.lelli@...hat.com>,
 Dietmar Eggemann <dietmar.eggemann@....com>,
 Steven Rostedt <rostedt@...dmis.org>, Ben Segall <bsegall@...gle.com>,
 Mel Gorman <mgorman@...e.de>, Daniel Bristot de Oliveira
 <bristot@...hat.com>, Valentin Schneider <vschneid@...hat.com>,
 Alexander Atanasov <alexander.atanasov@...tuozzo.com>,
 linux-kernel@...r.kernel.org
Subject: Re: [PATCH] sched/fair: Do not scan non-movable tasks several times

On 18.12.2023 17:48, Vincent Guittot wrote:
> On Thu, 14 Dec 2023 at 15:42, Konstantin Khorenko
> <khorenko@...tuozzo.com> wrote:
>>
>> If busiest rq is small, nr_running < SCHED_NR_MIGRATE_BREAK and all
>> tasks are not movable, detach_tasks() should not iterate more than tasks
>> available in the busiest rq.
>>
>> Previously the (env->loop > env->loop_max) condition prevented us from
> 
> It's usually better to give the commit directly when we know it :
> Before commit : b0defa7ae03e ("sched/fair: Make sure to try to detach
> at least one movable task"),
> the (env->loop > env->loop_max) condition prevented us from ...

You are definitely right, added.

>> scanning non-movable tasks more than rq size times, but after we start
>> checking the LBF_ALL_PINNED flag, the "all tasks are not movable" case
>> is under threat.
>>
> 
> Fixes: b0defa7ae03e ("sched/fair: Make sure to try to detach at least
> one movable task")

Added this too, thank you.

>> Signed-off-by: Konstantin Khorenko <khorenko@...tuozzo.com>
>> ---
>>   kernel/sched/fair.c | 7 ++++++-
>>   1 file changed, 6 insertions(+), 1 deletion(-)
>>
>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>> index d7a3c63a2171..faa2a765e899 100644
>> --- a/kernel/sched/fair.c
>> +++ b/kernel/sched/fair.c
>> @@ -11219,7 +11219,6 @@ static int load_balance(int this_cpu, struct rq *this_rq,
>>                  .dst_rq         = this_rq,
>>                  .dst_grpmask    = group_balance_mask(sd->groups),
>>                  .idle           = idle,
>> -               .loop_break     = SCHED_NR_MIGRATE_BREAK,
>>                  .cpus           = cpus,
>>                  .fbq_type       = all,
>>                  .tasks          = LIST_HEAD_INIT(env.tasks),
>> @@ -11265,6 +11264,12 @@ static int load_balance(int this_cpu, struct rq *this_rq,
>>                   * correctly treated as an imbalance.
>>                   */
>>                  env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
>> +               /*
>> +                * If busiest rq is small, nr_running < SCHED_NR_MIGRATE_BREAK
>> +                * and all tasks are not movable, detach_tasks() should not
>> +                * iterate more than tasks available in rq.
>> +                */
>> +               env.loop_break = min(SCHED_NR_MIGRATE_BREAK, busiest->nr_running);
> 
> Should it be after more_balance: ?
> In case we do "more_balance:" on a new_dst_cpu and it ends up that
> finally there is no more movable task as we released the lock of
> busiest rq in the meantime ?

Well, both yes and no here.
If we simply move the env.loop_break assignment after "more_balance" label,
we will break the handling LBF_NEED_BREAK case.

But you are right, the new_dst_cpu case should also not just reset loop_break to 
SCHED_NR_MIGRATE_BREAK, but do the same min() calculation.

So i've added one more label before the env.loop_break = min(...) and used it in
the new_dst_cpu case.

> Also you can remove one more superfluous init of loop_break:

Yep, dropped it, thank you.

> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -11361,7 +11361,6 @@ static int load_balance(int this_cpu, struct
> rq *this_rq,
>                           */
>                          if (!cpumask_subset(cpus, env.dst_grpmask)) {
>                                  env.loop = 0;
> -                               env.loop_break = SCHED_NR_MIGRATE_BREAK;
>                                  goto redo;
>                          }
>                          goto out_all_pinned;
> 
>>
>>   more_balance:
>>                  rq_lock_irqsave(busiest, &rf);
>> --
>> 2.39.3
>>